DEPLOYDatabase

Brain

Gemini Robotics 2

Google DeepMind's second-generation VLA for robotics, released July 2026. First model to control a humanoid's legs, torso, arms, and hands under a single policy. Includes GR-ER 2.0 reasoning model with agentic 'think-before-acting' loop. Research partners include Apptronik and Boston Dynamics.

Foundation model · Maturity: Pilot · Closed

Machine-readable surfaces

Architecture

Gemini 2.0-based VLA with embodied reasoning, whole-body loco-manipulation, on-device variant

Key facts

Gemini Robotics 2 lineup (primary DeepMind)
Google DeepMind's July 30 2026 post introduces Gemini Robotics 2 as three models: Gemini Robotics 2 (VLA for motor control of full humanoids and bi-arm robots), Gemini Robotics ER 2 (embodied reasoning VLM for multi-step planning and multi-robot collaboration), and Gemini Robotics On-Device 2 (efficient local VLA). ER 2 is available on Google AI Studio and in private preview on Gemini Enterprise Agent Platform; VLA and On-Device models are early-access for partners ([Google DeepMind](https://deepmind.google/blog/gemini-robotics-2-brings-whole-body-intelligence-to-robots/))
Whole-body demos and partners (primary DeepMind)
The post says Gemini Robotics 2 can control entire humanoids (example: Apptronik Apollo 2 putting a watering can on a bottom shelf), five-fingered SharpaWave hands, and Franka Duo grippers, and that On-Device 2 can adapt to new bi-arm embodiments in a few hours with typically fewer than 200 examples. Partners thanked include Apptronik, Boston Dynamics, and Agile Robots. Treat demo success rates as author-reported research results ([Google DeepMind](https://deepmind.google/blog/gemini-robotics-2-brings-whole-body-intelligence-to-robots/))
Three-model robotics lineup
Google DeepMind describes Gemini Robotics 2 as a lineup spanning a VLA for motor control, an embodied-reasoning model for planning and collaboration, and an on-device VLA optimized for local robotic execution ([Gemini Robotics 2](https://deepmind.google/blog/gemini-robotics-2-brings-whole-body-intelligence-to-robots/)).
Whole-body control
Google DeepMind says the Gemini Robotics 2 VLA can control robots ranging from tabletop systems to full humanoids, including whole-body movement from feet to fingertips and dexterous hand and gripper actions ([Gemini Robotics 2 VLA](https://deepmind.google/models/gemini-robotics/vla/)).
Embodied reasoning distribution
Google DeepMind's model card describes Gemini Robotics ER 2 as a vision-language model for spatial, temporal, and physical reasoning, tool orchestration, and success detection for physical agents. It lists Google AI Studio and the Gemini API as distribution channels ([Gemini Robotics ER 2 model card](https://deepmind.google/models/model-cards/gemini-robotics-er-2/)).
Gemini Robotics 2 whole-body intelligence announcement (DeepMind)
Google DeepMind announced Gemini Robotics 2 as bringing whole-body intelligence to robots, describing models for advanced dexterity and multi-robot collaboration, with Gemini Robotics ER 2 framed for broader developer access and other variants for early-access partners. Treat availability and capability claims as company-stated ([Google DeepMind](https://deepmind.google/blog/gemini-robotics-2-brings-whole-body-intelligence-to-robots/)).
Embodied reasoning and on-device framing (DeepMind blog)
DeepMind's Gemini Robotics 2 materials distinguish embodied-reasoning and on-device pathways for robot control research and partner deployments. Treat product-split claims as company-stated ([Google DeepMind](https://deepmind.google/blog/gemini-robotics-2-brings-whole-body-intelligence-to-robots/)).

Developed by (1)

Common questions

What is Gemini Robotics 2?
Google DeepMind's second-generation VLA for robotics, released July 2026. First model to control a humanoid's legs, torso, arms, and hands under a single policy. Includes GR-ER 2.0 reasoning model with agentic 'think-before-acting' loop. Research partners include Apptronik and Boston Dynamics.
Who developed Gemini Robotics 2?
Gemini Robotics 2 is credited to Google DeepMind on the DEPLOY registry. Each developer attribution is verified via primary sources.
Is Gemini Robotics 2 open source?
No. Gemini Robotics 2 is recorded as proprietary on the DEPLOY registry. Model weights and source are not publicly available.
What type of AI is Gemini Robotics 2?
Gemini Robotics 2 is a foundation model, built on a Gemini 2.0-based VLA with embodied reasoning, whole-body loco-manipulation, on-device variant architecture on the DEPLOY registry.
What is Gemini Robotics 2's maturity stage?
Gemini Robotics 2 is at the pilot stage on the DEPLOY maturity ladder. Pilot stage means at least one named-customer trial deployment is verified.
Which robots run on Gemini Robotics 2?
No robot models on the DEPLOY registry are recorded as running Gemini Robotics 2. DEPLOY wires brain-to-model connections only when the wiring is verifiable from primary sources; absence may reflect pre-deployment or unverified manufacturer claims.

Sources (3)

  1. https://deepmind.google/blog/gemini-robotics-2-brings-whole-body-intelligence-to-robots/
  2. https://deepmind.google/models/gemini-robotics/vla/
  3. https://deepmind.google/models/model-cards/gemini-robotics-er-2/
Methodology: Verified · 3 sources (no primary) · last reviewed 2026-09-29

Verification posture

Verified

Low confidence

Review state

Stable

Last reviewed 2026-09-29

Maturity + lifecycle

Maturity stage: pilot

Sources by quality tier

3
unclassified
Unclassified source

The framework is documented at /methodology. Corrections at /corrections. Reviewer: DEPLOY editorial team.

Methodology surface for Gemini Robotics 2.

Canonical ID a4ec2f53-8f5b-4745-8ab6-1d855fc34e49